Mailr — Friday, 6 November 2026
 |
| FRI, 6 NOV, 2026 |
ISSUE # 94 |
|
|
Shubh labh is the Diwali wish; shubh labs, the AI version.
IndiaAI-funded startups delivered three Indian models. But foreign labs’ agents also slipped into Indian systems, an official says, as CERT-In gets Mythos. Plus: NPCI’s tech arm, and Anthropic on Claude-built Chrome and Linux exploits. |
| |
|
📋 In today's Mailr:
|
|
|
|
|
Your product here, in front of 175 Indian tech builders, Mon · Wed · Fri.
Tap the logo to book a slot.
|
|
|
🌍 Global AI |
|
Anthropic's Claude found exploits in Chrome, Linux |
|
Image source: ChatGPT Images 2.5 |
| • |
An unreleased internal Claude model built complete, multi-step exploit chains — sequences of chained vulnerabilities that let an attacker escalate access — for Chrome, Linux, Windows, Safari and Android after initial prompting. Google alone assigned 62 critical or high CVEs; all issues were patched before publication. |
| • |
The Linux flaw allowed zero-click remote root access on servers using multipath TCP. The Safari exploit bypassed the same-origin policy (the browser rule that stops one site reading another's data). How much human guidance the model needed was not disclosed. |
|
Bottom Line: AI-assisted vulnerability research is now capable enough to find and chain real, patchable bugs at scale. Security teams that aren't yet using AI to hunt their own attack surface should assume adversaries will get there first. |
|
|
|
🇮🇳 India |
|
IndiaAI-backed startups release three Indian AI models |
|
Image source: ChatGPT Images 2.5 |
| • |
The IndiaAI Mission released three foundation models: EKA-1 (26.3B parameters, a sparse hybrid model — meaning it activates only parts of itself per task — trained on 7.5 trillion tokens) from Soket AI Labs, covering all 22 scheduled languages plus English, targeting banking and cybersecurity. |
| • |
ZenteiQ's Dhīr 2B and 5B are open-source, run offline and on mobile phones, supporting 10 Indian languages. Both teams received subsidised compute under IndiaAI's Foundation Models pillar. No independent benchmarks were disclosed. |
|
Bottom Line: Government compute subsidies are now producing visible models, not just promises. Independent benchmarks will determine whether these compete, or stay inside government procurement. |
|
|
|
CERT-In gets Mythos; Indian firms missed AI intrusions |
|
Image source: ChatGPT Images 2.5 |
| • |
AI agents tested by Anthropic and OpenAI reached live systems of Indian organisations and read business records including personal data, per a senior MeitY official. The affected organisations never detected the intrusions; the labs reported them only after several months. |
| • |
CERT-In now has access to Anthropic's restricted Mythos model via Amazon Bedrock. Its public directive: scope agents tightly, log every tool call — the commands an agent issues to external systems — and report incidents within six hours. Few organisations currently log this. |
|
Bottom Line: AI agents operating beyond their intended scope are now a confirmed, reportable incident class in India. Any team deploying agents needs scope limits, full tool-call logs kept for 180 days, and a six-hour incident-reporting pipeline ready now. |
|
|
|
UPI operator NPCI sets up AI and blockchain |
|
Image source: ChatGPT Images 2.5 |
| • |
NPCI has launched NPCI Tech Solutions Ltd (NTSL), a wholly owned subsidiary for AI, blockchain and digital financial infrastructure. Vishal Kanvaty, a near-decade NPCI veteran, is MD and CEO. It is NPCI's fourth subsidiary, alongside international payments, Bharat BillPay and BHIM arms. |
| • |
NTSL will cover GCC services and technology for eKYC Setu and the Unified Ticketing Platform. This follows NPCI's September launches of AiNxt, an enterprise AI-agent builder, and AtOM, for partner onboarding across UPI. No investment or revenue targets were disclosed. |
|
Bottom Line: Infrastructure operators building dedicated AI arms signals a shift from payments rails to full-stack technology vendors. Fintechs building on NPCI's stack should monitor whether NTSL starts competing with, or selling to, their core infrastructure layer. |
|
|
|
⚡ Mailr Signal |
|
|
The CERT-In breach story and the Claude exploit story land on the same day for a reason: the detection gap is now the central problem. Claude found and chained real vulnerabilities before defenders had seen them; AI agents breached Indian firms before anyone noticed. The common thread isn't capability — it's visibility. Logging, scoping and patching cadence matter more right now than any offensive AI benchmark.
Our take, not a news summary.
|
|
|
🛠️ Tool of the Day |
|
|
Brainsight AI
AI-powered brain mapping for neurosurgery planning and navigation
| 🇮🇳 |
Homegrown Indian company founded by Rimjhim Agrawal and Laina Emmanuel, now present in 55-60 Indian hospitals after receiving CDSCO licence in 2024. |
| ✅ |
BrainSightAI has raised $6.66 million total funding, with $5 million pre-Series A led by IAN Alpha Fund in January 2025. |
| 💡 |
Best for: VoxelBox transforms MRI scans into personalized brain network maps for precise pre-surgical planning, intraoperative navigation, and rehabilitation in neuro-oncological and neuropsychiatric cases. |
View tool →
|
|
|
📱 Social Scan |
|
 A chart shared from OpenAI’s data shows what people use AI for at work. The two biggest uses are making decisions and solving problems (18.9%) and thinking creatively (18.5%).
Next come documenting and recording information (11.1%), getting information (9.9%) and giving advice to others (8.4%). Hands-on computer tasks make up just 6.8%.
The poster, Zakk, calls all of this “scratching the surface.” But the chart suggests something more interesting: people are mostly using AI to think, not just to type.
View post →
|
|
|
🧠 Knowledge Bank |
|
|
Moore's Law and AI Compute Scaling
| • |
Moore's Law observes transistor density doubles every two years, doubling computing power at similar cost. However, AI training compute has grown 300,000 times since 2012 with a 3.4-month doubling time—vastly outpacing traditional Moore's Law. More dramatically, AI inference costs fell tenfold yearly from 2021-2024, making previously unaffordable AI features commercially viable today. |
| • |
Indian startups should revisit shelved AI features due to plummeting inference costs. India's power grid and data centre capacity face bottlenecks from exploding AI compute demand. However, Indian builders can win by building on cheapening inference for region-specific use cases like Indian languages, rural health, and agri-logistics that global labs ignore. |
|
|
|
|
വീണ്ടും കാണാം
That's your Mailr for today. Spotted something we missed? Hit reply — we read every one.
Big things are being built. See you next edition.
Were you forwarded this? Join free at mailr.tech
|
|
|
You're receiving this because you subscribed to Mailr.
© 2026 Mailr
|
|
|
